Published inTowards Data ScienceCODIR: Train smaller, faster NLP modelsA unique approach to knowledge distillationJan 4, 2021Jan 4, 2021
Published inTowards Data ScienceSpanFact: Fix your factually incorrect summariesA suite of models that fixes inaccuracies in summaries of large textDec 16, 2020Dec 16, 2020
Published inTowards Data ScienceHERO: You’ll never have to watch long videos againA model that understands a video and its subtitles to help you speed through itSep 17, 2020Sep 17, 2020
Published inThe StartupFILTER: Understand Foreign Languages BetterA technique that promotes a deeper understanding and linkage between the same text in different languagesSep 16, 2020Sep 16, 2020
Published inTowards Data ScienceDiscoBERT: A BERT that shortens your reading timeA Discourse Aware BERT that improves summary generationAug 22, 2020Aug 22, 2020
Published inThe StartupDistill-BERT: Using BERT for Smarter Text GenerationGenerate better stories, essays, answers and more using BERT’s knowledgeAug 7, 2020Aug 7, 2020
Published inTowards Data ScienceVILLA: A Generic Adversarial Training technique for Vision-and-LanguageThe first multimodal adversarial training technique for image and text modelsJul 13, 2020Jul 13, 2020
Published inThe StartupBachGAN: Let your imagination run freeCreate real images with just boxes!May 15, 2020May 15, 2020
Published inThe StartupVIOLIN: Do you really understand videos?A new task and dataset that requires multi-modal understanding of videos and textMay 3, 2020May 3, 2020
Published inTowards Data ScienceUNITERLearning a joint representation of image and text that everything can useMar 5, 20201Mar 5, 20201